Financial application of genetic programming

Sammanfattning: With the increasing speeds of modern processors the possibility of using genetic programming for problems with a huge amount of data has become feasible. One area where people over the course of time have been interested in looking for pattern is in the financial markets. Due to the nature of financial markets it is very hard to find patterns with traditional techniques. It is hoped that genetic programming can find these patterns that can’t be found in other ways, if they exist. This report studies genetic programming and a system called TSL that creates trading models with the help of genetic programming. TSL is built on a genetic programming software called Discipulus which is a very fast machine code based regression and classification tool. The first step before a run can take place is to collect the financial data that the models will be built from. During this work the data has been taken from TradeStation which is a system used for analyzing and trading the financial markets. After this is done TSL must be set up for the run. It has a lot of different parameters for the user to configure. When the run is over some of the models are saved and these can be tested in TradeStation to see their performance on another time period. If it gives a satisfactory result the models can be used for live trading. During the work I have focused on two futures contracts, Standard & Poor’s 500 E-mini contract and the British Pound contract. On these instruments extensive testing has been made but I have not been able to find any models that return risk adjusted excess returns during my work. There is a possibility thou that such systems actually has been produced during the evolution but due to flaws in the saving mechanism in TSL some of the most promising looking models have not been saved.